You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hive.apache.org by Dima Machlin <Di...@pursway.com> on 2014/01/22 14:21:00 UTC

HiverServer2 Logging

Hey,
We are considering to start using HiveServer2 and its seems that the logging capabilities are really limited as the server runs on a single port.
I'll explain the problem more accurately :
Today, we are using HiveServer where each user gets a specific port (or a set of ports) then, we start X hiveservers , 1 for each port in the following manner

nohup hive --service hiveserver -p $2 -v --hiveconf hive.root.logger=INFO,console >> ~/hive.$2 2>&1 &
For example :
nohup hive --service hiveserver -p 12354 -v --hiveconf hive.root.logger=INFO,console >> ~/hive.12354 2>&1 &

This way, hiveserver runs in the background, the logging is redirected to the "console" but then its redirected to a file.
So each port has its own log file, assuming there are no concurrent executions on the same port, each file is a sequential execution log.

This is really convenient to debugging executions on the HiveServer as the full stacktrace isn't passed through the JDBC and mostly we get a "MapredTask" error
The full errors are found in the logs and debugging is easy.

Now in the HiveServer2, I couldn't find any way to separate  the logging per session.
In fact i see that there is a Jira<https://issues.apache.org/jira/browse/HIVE-5924> open for this issue.

Is there any work around for this? I want each session / connection in its own log file, higher resolution is also acceptable (like a log per session)
In fact, looking at hive-exec-log4j.properties it seems that logging is per query :
hive.log.file=${hive.query.id}.log
But this is a different log full with counter data / task progress when i'm looking for the standard logging as seem in the CLI.
I think that this is defined by hive.log.file in hive-log4j.properties which is set to :
hive.log.file=hive.log
Trying to set it also as the previous doesn't help as ${hive.query.id} resolves to nothing.

Any advice appreciated.

-----------------------------------------------
[cid:image001.jpg@01CE92B5.CB034C90]
Dima Machlin, Big Data Architect
Pursway.com<http://www.pursway.com/>